AI & ML Research 24 Hours

Last updated: May 21, 2026, 8:39 AM ET

AI Agents: Cost and Safety

Running AI agents can become prohibitively expensive without clear strategies for planning, skill coverage, and budget management, so practitioners are applying operations research and data science techniques to optimize agent workflows and cap costs. Safely deploying coding agents in production demands sandboxing, permission scoping, and human-in-the-loop checkpoints to prevent unintended code execution, limiting adoption despite productivity gains.

Synthetic Surveys

A new unlearning approach addresses mode collapse in synthetic survey replies generated by large language models, enabling LLMs to produce more diverse responses that better replicate human respondent distributions.